B&LdJ1 Theoretical Issues in Psychology Philosophy of Science and Philosophy of Mind for Psychologists.

B&LdJ 1

Theoretical Issues in Psychology

Philosophy of Scienceand

Philosophy of Mindfor

Psychologists

B&LdJ 2

Chapter 8Modern approaches to mind (2)

• Neural networks and connectionism• ‘Classical’ versus connectionist architecture• The third contender: dynamicism• Is there need for ‘mental representations’?• Naturalism, reductionism and folk psychology

B&LdJ 3

Neurophilosophy (the Churchlands)

Contrary to Fodor’s ideas:• For answers to philosophical-psychological questions turn to the neurosciences (empirical knowledge - naturalism).• No autonomy for psychology.• Brain-based view of mind: ‘mind is brain’ (but no mind-brain identity).• Eliminate folk-psychological concepts.• Against sentential (propositional) LOT-view of knowledge.• No solipsism, but evolutionary adaptation.

B&LdJ 4

Model of a neural network

• Set of nodes and connections between them.• Activation spreads through the network.• Giving weights to the nodes.• Three layers: input, output, midden layer of hidden nodes.• Network learns by adjustments of weights.• According to some learning rule (‘Hebb rule’).• Does the training by itself, is tuned to the environment (isnot pre-programmed).• Maintains fault tolerance, graceful degradation (functional persistence), and satisfies soft constraints (all unlike programs).• Knowledge/representations is/are distributed over many connections.• Network is model of human mind: ‘connectionism’.

B&LdJ 5

Working of a (quasi-neural) network(here, a network of a submarine, learning sonar signals)

B&LdJ 6

Activation space and prototypes

• Learning to recognize is forming a prototype, a hotspot in a vector or activation space.• A concept is a prototype (not a symbol string).

B&LdJ 7

Connectionism

• (Quasi-)neural networks:• ‘neural’ patterns of activation (versus symbols and

inborn rules);

• online selforganisation (versus offline programs);• support cognitive tasks.

• Basic cognitive processes (formation of representations) are patterns of activation (not manipulation of symbols).

• Cognition is basically pattern recognition. • Neurophysiology, c.q. neural networks, explains

cognition: ‘cognitive neuroscience’, cognitive theories which are neurological plausible and naturalized:• no functionalism;• no autonomy for psychology.

B&LdJ 8

• Human brain has: 100 billion neurons = 100.000.000.000 (1010).

• Every neuron has: synaptic connections with 10.000 (104) other neurons.

• In a human brain: 1014 synaptic connections.

• Weight of each connection can have a value of 1 out of 10.

• Thus human brain can contain: 101014 = 10100.000.000.000.000

cognitive configurations(cfr. total elementary particles in universe estimated 1087).

Representational capacity of human brain

B&LdJ 9

Symbolic versus connectionistic systems

Fodor: thinking is characterized by productivity, systematicity,i.e. a continuous recombination of discrete symbols: compositionality, like building sentences with words: requires formal structure (networks lack structure and are not strong enough to simulate cognition; leaves you with nothing more than a plain associationpsychology);so, LOT is ‘the only game in town’: thinking occurs in a formal language; the only explanation of structure.

Connectionists (a.o. Smolensky): thinking is network activation, compositionality is a by-product of networks: ‘functional compositionality’: doesn’t need a symbolic architecture; productivity of language is not the only possible productivity (Churchland).

B&LdJ 10

Third Contender: Dynamism: ‘mobots’ and dynamic systems

• Dynamic coupling of organism and environment.

• Activity in environment (no inner representational

domain).

• Online interaction (no innate structure and programs).

• Evolution and time (no static representations).

• Adaptation.

• Dynamic system (no representational structure: LOT or

patterns of activation).

• Cognition is like playing tennis, rather than chess.

B&LdJ 11

Dynamism: mobots

Mobots : internal representations

and computations redundant.

Rodney Brooks: robot with

response systems, direct

interaction with environment,

without central representations.

B&LdJ 12

Dynamism: Watt governor

Cognition is on-line real-time interaction with the environment – Watt governor, continuous following and control of behavior and environment; reciprocal causality.

B&LdJ 13

Mental representations redundant?

The absent and the abstract:• there are ‘representation-hungry’ situations.• that require higher cognitive functions like abstract thinking, imagination and reflection.

• Direct coupling mainly in sensori-motor functions.• External symbols for higher cognitive tasks (planning, abstraction).• ‘Active externalism’ (extended mind, Ch 9.4): for some cognitive tasks we use external instruments, e.g., paper and pencil, graphical devices; books, internet.• ‘Leaky cognition’: brain and environment cannot be separated.

Andy Clark (1997, 2003)

B&LdJ

Naturalism and neurophilosophy Churchland

Naturalism: representation is biological phenomenon, product

of evolution.

Cognition is pattern recognition, not symbol manipulation.

Language appears late in evolution, therefore no LOT.

Functionalism is a conservative ‘cheap# explanation.

Folk psychology (beliefs and desires) stagnating program,

isolated from scientific progress.

Folk psychology eliminated and replaced by neuro-speak.

14

B&LdJ 15

Naturalistic (neuroscientific) explanations of:

• knowledge as coded in connection weights;

• representation and intentionality as processes in the

brain;

• sense of self;

• morality as cognitive skill; role of oxytocine in building trust

and love.

Churchland: Naturalism and neurophilosophy

B&LdJ 16

Eliminativism and folk psychology

Folk psychology (‘belief-desire’) is a kind of theory, explains behavior (Fodor + Churchland).

Fodor: in principle correct theory:• beliefs and desires exist really as symbol structures in LOT;• are causes of behavior;• folk ps. vindicated by CTM;• intentional laws, generalisations, and predictions indispensable: folk psychology is successful predictor;•intentional realism.

B&LdJ 17

Eliminativism and folk psychology

• Churchland: folk psychology obsolete theory:• stagnating program (Lakatos);• no progress, no connections with current development in

science;• replace by neuroscience;• eliminativism.

• Clark, Dennett: folk psychology is not a scientific theory, • beliefs and desires only descriptive (‘intentional stance’), not internal causes (‘design stance’), not literally true;• different explanatory aims; different level from neuroscientific explanation;• instrumentalism.

B&LdJ 18

CTM Connectionism Dynamic systemsFormal, syntactical rules, symbols

Weights and activation patterns

Coupled co-evolving systems, developing over time

Preprogrammed, no real development

Self-organisation, learning through adapting weights

Evolving through state space, circular causality, continuous adaptation

Brittle program rules Graceful degradation under damage

Smooth mutual adaptation

Structured, language-like architecture, concatenating discrete symbols

‘Associationism’ Development in time

Productivity and systematicity through compositional architecture

Functional compositionality

Trajectory through state space

Functionalism, autonomy for psychology

Reductionist, (more or less) brain-like cognition

Emergent properties of organism-environment system, and development

Representations are symbolic structures

Representations are activation patterns

No representations needed

Solipsism, self-contained mind

Representations are products of interaction with environment

Body mind and world part of a single system